I understood that those are improvements on the way cycles X interfaces with viewport. Task sorting and priority stuff rather than brute computation enhancements.
Real gain in render time so far is only with (single) nvidia GPU rendering
CPU rendering on Linux is significantly faster both in viewport and final render.
Hi, update from Sergey, there is now a Full path tracing report if you start with --debug-cycles switch.
https://developer.blender.org/D11105
Result:
Full path tracing report
Path tracing on: GeForce RTX 2060 [CUDA_GeForce RTX 2060_0000:02:00_OptiX]
Render Scheduler Summary
Mode: Background
Resolution: 1344x756
Adaptive sampling:
Use: True
Step: 16
Min Samples: 22
Threshold: 0.001953
Denoiser:
Use: False
Time (in seconds):
Wall Average
Path Tracing 25.361249 0.049534
Adaptive Filter 0.053353 0.000108
Display Update 0.417720 0.041772
Total: 25.832322
I0428 12:49:05.209913 22631 blender_session.cpp:541] Total render time: 26.9057
I0428 12:49:05.209933 22631 blender_session.cpp:542] Render time (without synchronization): 25.7606
I0428 12:49:05.231812 22631 device_impl.cpp:85] Closed database: "/var/tmp/OptixCache_pepo/cache7.db"
I0428 12:49:05.245733 22631 device_impl.cpp:85] Cache data size: "285.5 MiB"
Cheers, mib
He refers to regular Cycles vs. Cycles-X comparison. You wonāt get 2x faster render times with CPU and Cycles-X in that comparison.
Yes that is what I meant. By the way I forgot to write an āXā
Anyways, I always refer to what I saw in the Live presentation AND what I understand from it. I absolutely donāt know anything of the code involved, so I might also say incorrect things, and - in case - I apologize in advance
A simple scene with glass and hdri go from 1:57 to 1:20
My initial impressions
- A slightly complex scene of mine opened with no errors, that is good.
- Apparently scenes will render properly as long as thereās no AO/Bevel shading and volumetrics (as they arenāt in yet), good.
- This is in dire need of being synched with master, as there are no geometry nodes in the first build.
- Rendering appears faster with the CPU, meaning my Ryzen+ 2700X can be stretched out a bit further, it takes the pressure off of upgrading the whole box sooner rather than later, that is nice.
- No tiles in rendering is a blast from the past, but in this case it is not a bad thing at all.
It looks quite promising, good job Brecht and Sergey.
Definitely not as impressive as the GPU rendering improvements, but renders are actually about 25% faster for me on the CPU (on Mac).
Any increase in render speed is nice and the results so far look very promising from this aspect - but iām more excited to see what improvements to the actual render engineās capabilities fall out of this project.
As Iāve said, my preferred priorities are multi-device and volumetrics, though I guess Iām probably in a minority using multiple-GPUs, and volumetrics can be faked. Nevertheless, I have high hopes.
mmmā¦ I think as soon as bevel is in Iāll start using Cycles X for production!
Iāve just made a new Cycles-X build, after it was updated by Brecht and Sergey on April 28th. It can be downloaded from the Experimental Branches page.
Just few test renders with CyclesX.
Most of all impressed with responsiveness during GPU viewport rendering.
At least twice faster out of a box!
Did a test last night on the Blender monster sss demo file the one with the big pink blob monster. CyclesX was more than twice as fast than the standard Cycles. And slightly brighter reflections and causticsā¦ Absolute win!
Hi, developement goes further:
https://developer.blender.org/rB2bed80609a87acd6886aace21cfebf86c31c3d84
3-12 % faster render is anounced.
Cheers, mib
EDIT: I cant get any performance change but was only testing Blob (Metin Seven) and own file.
it still says April 23 on that page
You must be using Mac. It is common for these builds to take a little longer.
Cycles-X: Faster & Brighterā¢